Multimodal Conversation on LLaVA Bench
93.1LLaVA Bench ScoreGPT-4V (2023.11.06)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4V (2023.11.06)2024.08 | 93.1 | — | — | — | |
| MiniCPM-Llama3-V 2.5Size=8.5B2024.08 | 86.7 | — | — | — | |
| Baseline: all KVVLM Backbone=Qwen2.5-VL-7B-Instruct2026.05 | 85.9 | — | — | — | |
| KVCapsuleVLM Backbone=Qwen2.5-VL-7B-Instruct2026.05 | 83.7 | — | — | — | |
| CogVLM2-19B-ChatSize=19B2024.08 | 83 | — | — | — | |
| Cambrian-34BSize=34B2024.08 | 82 | — | — | — | |
| LLaVA-NeXT-Yi-34BSize=34B2024.08 | 81.8 | — | — | — | |
| Baseline: all KVVLM Backbone=Qwen2-VL-7B-Instruct2026.05 | 80.8 | — | — | — | |
| SparseVLMsVLM Backbone=Qwen2.5-VL-7B-Instruct2026.05 | 80.2 | — | — | — | |
| LLaVA-NeXT-Llama3-8BSize=8.4B2024.08 | 80.1 | — | — | — | |
| Gemini Pro2024.08 | 79.9 | — | — | — | |
| PrumergeVLM Backbone=Qwen2.5-VL-7B-Instruct2026.05 | 78.2 | — | — | — | |
| DeepSeek-VL-7BSize=7.3B2024.08 | 77.8 | — | — | — | |
| SparseVLMsVLM Backbone=Qwen2-VL-7B-Instruct2026.05 | 75.3 | — | — | — | |
| KVCapsuleVLM Backbone=Qwen2-VL-7B-Instruct2026.05 | 74.8 | — | — | — | |
| Baseline: all KVVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 74.8 | — | — | — | |
| PrumergeVLM Backbone=Qwen2-VL-7B-Instruct2026.05 | 74.7 | — | — | — | |
| Claude 3 Opus2024.08 | 73.9 | — | — | — | |
| CogVLM-ChatSize=17B2024.08 | 73.9 | — | — | — | |
| KVCapsuleVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 73.8 | — | — | — | |
| PrumergeVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 73.4 | — | — | — | |
| SparseVLMsVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 72 | — | — | — | |
| Cambrian-8BSize=8.3B2024.08 | 71 | — | — | — | |
| XTuner-Llama-3-8B-v1.1Size=8.4B2024.08 | 69.2 | — | — | — | |
| MiniCPM-V 2.0Size=2.8B2024.08 | 69.2 | — | — | — | |
| Baseline: all KVVLM Backbone=llava-v1.6-mistral-7b-hf2026.05 | 67.8 | — | — | — | |
| KVCapsuleVLM Backbone=llava-v1.6-mistral-7b-hf2026.05 | 67.8 | — | — | — | |
| SparseVLMsVLM Backbone=llava-v1.6-mistral-7b-hf2026.05 | 67.8 | — | — | — | |
| Qwen-VL-ChatSize=9.6B2024.08 | 67.7 | — | — | — | |
| FastVVLM Backbone=llava-v1.6-mistral-7b-hf2026.05 | 67.7 | — | — | — | |
| PrumergeVLM Backbone=llava-v1.6-mistral-7b-hf2026.05 | 66.5 | — | — | — | |
| Baseline: all KVVLM Backbone=llama3-llava-next-8b-hf2026.05 | 66.3 | — | — | — | |
| KVCapsuleVLM Backbone=llama3-llava-next-8b-hf2026.05 | 66 | — | — | — | |
| SparseVLMsVLM Backbone=llama3-llava-next-8b-hf2026.05 | 65.2 | — | — | — | |
| PrumergeVLM Backbone=llama3-llava-next-8b-hf2026.05 | 65.2 | — | — | — | |
| FastVVLM Backbone=llama3-llava-next-8b-hf2026.05 | 64.3 | — | — | — | |
| Phi-3-vision-128k-instructSize=4.2B2024.08 | 64.2 | — | — | — | |
| Yi-VL-34BSize=34B2024.08 | 62.3 | — | — | — | |
| Bunny-Llama-3-8BSize=8.4B2024.08 | 61.2 | — | — | — | |
| FastVVLM Backbone=Qwen2-VL-7B-Instruct2026.05 | 55.5 | — | — | — | |
| Yi-VL-6BSize=6.7B2024.08 | 51.9 | — | — | — | |
| MiniCPM-V 1.0Size=2.8B2024.08 | 51.3 | — | — | — | |
| DeepSeek-VL-1.3BSize=1.7B2024.08 | 51.1 | — | — | — | |
| Idefics2Size=8.0B2024.08 | 49.1 | — | — | — | |
| FastVVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 22.3 | — | — | — | |
| FastVVLM Backbone=Qwen2.5-VL-7B-Instruct2026.05 | 12.9 | — | — | — | |
| GPT-4V2023.12 | — | 96 | 102.5 | 106.7 | |
| InstructBLIP2023.12 | — | 83.2 | 67.6 | 90.6 | |
| LLaVA2023.12 | — | 85.4 | 74.3 | 96.3 | |
| LLaVA 1.52023.12 | — | 81.6 | 75.5 | 95.2 | |
| LLaVA-RLHF2023.12 | — | 93.8 | 74.3 | 111.4 | |
| LRV2023.12 | — | 61.7 | 47.3 | 55 | |
| Muffin2023.12 | — | 89.3 | 79.7 | 97.7 | |
| Qwen-VL-Chat2023.12 | — | 81.9 | 77.1 | 92.3 | |
| RLHF-VBackbone=Vicuna v1.0 13B, Visual Module=BEiT-3, Resolution=4482023.12 | — | 93.1 | 75.3 | 91.6 |